📄 Armenian AutoEpiDoc: Automated Extraction and Encoding of Armenian Inscriptions into EpiDoc TEI/XML

Summary

Armenian epigraphy is well documented in printed scholarship such as the multi-volume Divan Hay Vimagrut'yan (DHV). However, it has almost no machine-readable editions, so these inscriptions are missing from TEI/EpiDoc-based infrastructures like EAGLE and FAIR Epigraphy. Encoding EpiDoc by hand demands both XML skills and knowledge of Armenian epigraphic conventions, which makes large-scale digitization impractical.

The authors present Armenian AutoEpiDoc, which they describe as the first system to turn expert-validated Armenian inscription records into EpiDoc-compliant TEI/XML. It does not rely on OCR. Instead, it takes curated CSV spreadsheets and processes them in three steps:

• loads them into a normalized MySQL database,
• builds TEI authority files linked to external vocabularies,
• serializes each record to XML using rule-based, schema-aware mappings, including a parser for Armenian diplomatic transcription conventions.

In pilot comparisons, manual encoding took tens of minutes per inscription, while the system produced a first draft in seconds that needed only targeted correction. Specialists confirmed that Armenian conventions are preserved. The tool lays groundwork for a first standards-aligned digital corpus of Armenian inscriptions.

Key points

• Pipeline: The Python ELT workflow has three modules: csv_to_mysql.py for ingestion, header normalization and a star schema, mysql_to_authority_list.py for authority files, and mysql_to_epidoc.py for XPath-based TEI serialization.

• Authority files: These cover Places, Monuments, Materials and Object Types. Place hierarchies are rebuilt via <note type="relation">, for example for Artsakh. Materials and object types are aligned with the Getty AAT and EAGLE vocabularies in a <standOff> section.

• Transcription parsing: Simplified ASCII transcriptions are mapped to EpiDoc elements:

  • <expan>/<abbr>/<ex> for abbreviations,
  • <num> for letter-numerals,
  • <hi>/<join> for ligatures and overlines.

  Damaged numerals and ambiguous ligatures are flagged for expert review rather than forced into a reading.

• Dates and language: Each date is recorded using two calendar systems, the Armenian Era and the Gregorian, and all metadata is bilingual (hy/en).

• Context and availability: Over 80% of Armenian inscriptions are located outside the modern Republic of Armenia. The system is in active use, for example in the Artsakh Epigraphy Atlas, and the code is released at github.com/dhlab-epfl/autoepidoc.